# MAIAR

A modular agent framework built around triggers, decision-making and pluggable actions, letting developers compose an agent that reacts to inputs and executes operations without writing a fixed workflow.

MAIAR is a composable, plugin-based framework for building AI agents. It treats an agent as three steps: data ingestion and triggers, decision-making, and action execution, and lets developers define triggers and actions as standalone plugins while the core runtime handles decision-making dynamically.

A generated project starts with the OpenAI model provider by default, SQLite-based conversation memory, text and image generation, console logging and logging over a WebSocket for remote monitoring. Models, providers and memory storage can be swapped, and pipes chain multiple steps together into more complex behavior.

It is written in TypeScript, is installed with a scaffolding command, and runs as a local server that an agent can be queried against over HTTP.

Website: https://maiar.dev